Underway xCO2 Measurements
The R/V Thomas G. Thompson departed Suva, Fiji, on October 5, 1993, for WOCE leg P10. C. Sabine started the Princeton underway pCO2 system the following day and provided necessary maintenance until the system was shut down the day before entering Yokohama, Japan, on November 10, 1993. Major problems with the gas selection valve were encountered on the cruise, resulting in a gap in data collection between 14º and 29º N. On average, one set of surface water and atmospheric CO2 mole fraction (xCO2) measurements was collected by the Princeton underway system approximately every 5 min while the system was operating normally. The sample locations along the cruise track are shown in Fig. 6.
Methods for Measurement and Computation
The Princeton underway CO2 system uses a rotating disk equilibrator design with an infrared detector that has been shown to provide stable, consistent results with minimal attention by the operator (Sabine and Key 1996, 1997). The equilibrator is a modified disk-stripper design that was found to be very efficient at removing radon from seawater (Schink et al. 1970). The components of the system are linked to a computer and sample analysis is fully automated. The primary advantages of this system are the rapid response time of the equilibrator and the low level of expertise necessary to maintain the system relative to a gas chromatographic detector design. The gain in equilibrator response time sacrifices simplicity relative to the shower-head equilibrator, since the disk equilibrator has moving parts in addition to the air-circulation pump present on all seagoing pCO2 instruments. On the other hand, the disk equilibrator design is not sensitive to changes in water pressure from the bow pump.
Figure 7 shows the major components of the Princeton system. Uncontaminated water from the ship's bow pump flows through the lower half of a disk equilibrator at approximately 18 L/min. CO2 in the water equilibrates with recirculated air in the top half of the chamber. Equilibrator air, air pumped from the ship's bow/stern (depending on wind direction), and four standard gases (reference, low, mid, and high) are plumbed into a computer-controlled gas-sampling valve that determines which gas is directed to the detector. The mole fraction of the sample gas is determined by a Li-Cor 6251 nondispersive dual-beam infrared detector. During calibration, the instrument measures the concentrations of four standards that have a range of CO2 concentrations in air, normalizes the detector voltage to temperature and pressure, and fits the results with a third-order polynomial. A full description of the system is presented by Sabine and Key (1997) with a few minor differences as noted in the following sections.
The system used on this cruise was a predecessor of the system described by Sabine and Key (1997) used during the WOCE Indian Ocean cruises. The primary physical difference between the systems was the lack of thermal control on the detector rack on the earlier system. Although normalizing the detector response to the measured temperature removed most of the short-term variability in the detector response to standards, the remaining variability was still correlated with detector temperature. This variability was removed in the calibration routine (see discussion bellow), but it was decided to minimize this complicating effect in later versions of the system by enclosing the detector and associated plumbing in a temperature- controlled box.
The working standard gas concentrations were calibrated against primary CO2 standards, provided by P. Tans (NOAA/CMDL), in the laboratory using the seagoing detector prior to the cruise. Working standards were a mixture of CO2 in artificial air prepared by Scott Specialty Gases. Multiple measurements of the working standards (0.00, 282.51, 349.77, and 400.70 ppm CO2) were made with the detector calibrated against the primary standards. Measurement precision was better than 0.1 ppm on all standards.
Equilibrator temperature was monitored by a Rosemont ultralinear platinum resistance thermometer (PRT). The PRT was calibrated in the laboratory against a National Institute of Standards and Technology (NIST) traceable mercury thermometer. Estimated accuracy was ±0.01ºC on the ITS90 scale.
Temperature readings from the Li-Cor detector were not explicitly calibrated for this survey because the final results are only a function of the relative changes in temperature between the standard gases and the sample.
The sensor used to monitor the system pressure (Setra Systems Inc.) was factory- calibrated against NIST-traceable primary standards prior to the cruise. Estimated accuracy was ±0.05%.
All system inputs were read into the computer as voltages using a Keithly A/D board. Accuracy of the board's readings was confirmed with a Fluke model 8840A 5-digit voltmeter prior to the cruise. The resolution of the readings was a function of the voltage range being measured, but in all cases was at least an order of magnitude smaller than the estimated precision of the measurement.
Both the sea surface temperature and salinity values were calibrated against the WOCE preliminary surface bottle values at each station. Although the exact trip time is not generally recorded in the WOCE ".SEA" files, the ".SUM" files do record the beginning and ending times of each cast. Since the Niskin bottles were tripped on the upcast, the surface bottle was tripped immediately before the rosette was brought aboard and the cast was completed. The end time for the cast was taken, therefore, as the trip time for the surface bottle at each station. The surface station data were then tied to the underway data by calculating the mean and median values of the underway data for the 15 min prior to the recorded cast end time. Although the ship was not underway while the cast was in progress, there was the potential that differences between the underway temperature readings and the discrete samples could have been real in very high gradient regions. Stations where the mean and median values were greater than 0.01 units apart were flagged, therefore, as questionable and not considered in the calibration fits.
The Princeton system normally operates automatically with a single microcomputer (80486 CPU) controlling sample selection, valve switching, and data logging. Figure 8 shows a typical record of detector voltages recorded for one full calibration and measurement cycle. The details of exactly how the system selects the standards and determines sample stability is described by Sabine and Key (1997). The primary difference between the operation of this version of the system and that described by Sabine and Key (1997) is the frequency at which the system sampled each gas. For this cruise, a full set of standards (reference, low, mid, and high standards) was analyzed every 3 h with a partial standard set (low, mid, and high standards) run every hour in-between full calibrations. An hourly calibration sequence was chosen to ensure that detector drift was captured, but this was found to be well in excess of what was necessary. The reference gas was analyzed only at 3 h intervals because the detector took a long time to stabilize with the 0-ppm CO2 reference and the 0-ppm detector readings were very stable. After calibration, the system alternately collected six marine air and six equilibrator sample gases until it was time for another calibration.
Listed below, in order of calculation, are the steps that were used to calibrate the results with the Princeton system.
- Average the readings (four per calibration) for the reference gas and each standard gas for each calibration run.
- Estimate the response for each gas as a function of time by calculating the set of linear regression lines that connect the estimated responses from the calibration runs. In other words, "connect the dots" generated by step 1 plotted as a function of time. Various smoothing curves could be used here, but this procedure yields the lowest uncertainty of any tried to date (possibly because of the short time scale correlation among the four results).
- Based on the four sets (one set for each standard gas) of regression lines generated by step 2, calculate the response for each standard gas at the time each equilibrator gas or bow air sample was measured.
- Use the results of step 3 with the detector response for the measurements to calculate the concentration of the unknown samples. Here it is assumed that the relationship between detector response and gas concentration follows a third-order polynomial; therefore, this step requires finding the real roots of a third-order polynomial for each unknown sample measurement.
The result obtained from these four steps is the xCO2 of the measured dry gas. This value can be corrected to pCO2 or fugacity of CO2 (fCO2) at in situ conditions. These adjustments have been described in great detail in DOE Handbook (DOE 1994). The CO2 concentrations reported in the final data tables were given at the measured equilibrator temperature (average 0.49 ± 0.1ºC greater than sea surface temperature) and were corrected to in situ temperature using the relationship of Weiss et al. (1982). In order to calculate xCO2 between surface water and the atmosphere, the atmospheric results were interpolated to the times surface water measurements were made. A separate file with the measured atmospheric values is also provided.
Precision and Accuracy
The primary calibration method for the system is periodic analysis of gas standards using known CO2 concentrations. The infrared detector response is slightly curvilinear (i.e., not straight) with respect to CO2 concentration in the sample gas path. Additionally, the detector has been found to have a slow drift over a period of several hours. Frequent calibration against standards can give an estimate of the analytical precision; however, this technique has the potential of systematic error with respect to accuracy. There is currently no known rigorous statistical test to determine optimal instrumental settings, or, for that matter, even to estimate the expected uncertainty of the results; however, it was attempted to estimate the precision and accuracy of these results with the best available information.
One estimate of precision was obtained by looking at the standard deviation of repeated measurements of the gas standards. The mean of the standard deviations determined for this instrument was 0.007 V, which was approximately equivalent to 0.9 ppm CO2. This can be interpreted as a crude estimate of the analytical precision if one assumes (1) that the precision obtained when analyzing sample gas is the same as for a standard and (2) that no additional uncertainty is incurred when interpolating calibration runs to the sampling time. This estimate assumes that no significant error was incurred in converting to in situ conditions. This assumption is reasonable with respect to precision, but not necessarily for accuracy. The primary correction to get xCO2 at in situ conditions was the warming of the water as it passed through the ship and equilibrator. A 1-degree increase in temperature results in roughly a 4% or ~14 ppm change in xCO2 (Takahashi et al. 1993; Copin-Montegut 1988; Weiss et al. 1982); therefore, both sea surface and equilibrator water temperatures must be known to an accuracy of ~.05ºC to calculate xCO2 values accurate to ±1 ppm. A second estimate of accuracy can be obtained by comparison with other systems. No other pCO2 systems were running on this cruise, but the Princeton system has given very consistent results with R. Weiss's gas- chromatograph-based/shower-type system when the two systems were run in parallel (Sabine and Key 1997). It was estimated that the accuracy of this system was ~1 ppm.
Two days after the cruise began, the underway CO2 system started to have problems with the Valco gas-sampling valve. Occasionally during position switching, the valve actuator would not stop, and the valve would continue to spin until the fuse blew or the power was switched off and back on again. By October 9 the problem had become more frequent. The valve was completely dismantled and cleaned. Valco technical support was contacted by phone. The problem was deduced to be in the Valco logic board, which could not be repaired at sea. Upon reassembly the valve seemed to work better. It was watched carefully and the spinning was corrected when necessary by switching the power to the valve off and on again. On the morning of October 14, the air flow in the bow air line had dropped substantially. The tubing was flushed with deionized water to remove any salt deposits in the line, and the valve port was cleaned with a cotton-tipped swab. The pump head was also replaced on the air cadet pump, and the flow returned to normal. On October 15, the Valco gas-sampling valve was disassembled and cleaned because sample gas flow was very low. Flow rate was much better after cleaning. On October 17 at 18:30 h, a problem was discovered with the drying tube while the system was running a standardization. The Nafion® tubing had apparently pulled out of the fitting. This problem could not have occurred too long before it was discovered because the sample was contaminated with 0-ppm gas, which was giving very unusual readings. The entire drying tube assembly was replaced, and the system was recalibrated. On October 18, the Valco spinning problem was too severe for the automated sampling program. The system was run using a program that required the operator to hit a key on the computer to change valve positions. This way, if the valve started spinning, it could be fixed right away. An attempted was made to run the system on the same schedule that the automated system was programmed to run. On October 23 at 22:30, the water pump failed, and the equilibrator was found drained. The system was switched to sample bow air while the water pump was repaired. On October 25, the bow and equilibrator gas flow rates were very low. The Valco was taken apart and cleaned. Upon reassembly, the actuator would do nothing but spin. Many unsuccessful attempts were made to repair the problem. The system was shut down until a solution could be found. On November 3, the ship's engine room crew managed to manufacture a handle that would allow the rotor to be manually rotated. The system was started again and run using the manual program for the remainder of the cruise.
Despite the many mechanical difficulties on this cruise, more than 11,600 surface seawater and 4,000 marine air measurements covering more than 50 degrees of latitude in the far western Pacific where collected. Figure 9 shows the surface water and atmospheric xCO2 concentrations as a function of the day of the year on the cruise. It is clear that the surface water CO2 concentration can vary significantly over relatively short time (space) scales. Values corrected to sea surface temperature ranged from a low of 314.8 ppm to a high of 389.9 ppm. Low xCO2 values were observed at the beginning of the cruise (near Fiji) and at the end of the cruise (near Japan). The highest xCO2 values were observed near the equator, crossed on day 289.